Early Detection and Diagnosis of Chronic Kidney Disease Based on Selected Predominant Features

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Nephrology

Background:

• Early detection of chronic diseases like chronic kidney disease (CKD) is vital to prevent severe outcomes.
• CKD diagnosis presents challenges due to high costs and difficulty in early identification.
• Timely medical decisions are essential in critical cases to avoid fatal consequences.

Purpose of the Study:

• To develop a prediction-based method for early detection and diagnosis of CKD patients.
• To enable a fast and accurate decision-making process for early-stage CKD.
• To reduce the cost of CKD diagnosis through efficient methods.

Main Methods:

• Developed a combination of data preprocessing and feature selection techniques.
• Trained and evaluated multiple prediction models including K-nearest neighbor (KNN), Support Vector Machine (SVM), Random Forest (RF), and Bagging.
• Utilized a processed dataset for model training and performance evaluation.

Main Results:

• All models demonstrated high reliability across accuracy, precision, sensitivity, F-measure, specificity, and AUC.
• K-nearest neighbor (KNN) achieved superior performance with 99.50% accuracy, 99.2% sensitivity, 100% precision, 98.7% specificity, and 99.6% for F-measure and AUC.
• The reduced feature set indicated that minimal clinical tests are sufficient for effective CKD detection.

Conclusions:

• The KNN model proved to be the best fit for CKD prediction compared to state-of-the-art methods.
• The developed prediction method facilitates rapid and precise early-stage CKD diagnosis.
• The study highlights the potential for significant cost reduction in CKD diagnosis.